Humanitarian Analytics
By Violet Chen
The technical session “Humanitarian Analytics” began on Sunday, October 20, at 11am. The session was chaired by Dr. Elisa Long, who is an Associate Professor of Decisions, Operations, and Technology Management at UCLA Anderson School of Management.
The first talk was given by Elnaz Kabir, a PhD student in Industrial and Operations Engineering from the University of Michigan, Ann Arbor. Elnaz’s talk was titled “Hurricanes, Power Outages, and Utility Response,” which is joint work with Chengwei Zhai and Seth Guikema. In this work, they developed a statistical model for predicting power outages resulted by hurricanes. In addition to introducing methodology, Elnaz shared how they transitioned research into practice: their forecasts are shared via web portals—a publicly available one and a secure one for utility companies. The fundamental goal is to use effective forecasting to inform better restoration planning, and Elnaz concluded by emphasizing that long-term close collaborations with utility companies are necessary for attaining such a goal.
Dr. Xiaodan Pan from Concordia University gave the second talk: “Product Availability, Consumer Stockpiling, and Hurricane Events: Empirical Evidence from Natural Experiments.” Dr. Pan is an Assistant Professor in Supply Chain and Business Technology Management. Consumer stockpiling is a common behavior observed in preparation for forecasted natural disasters, such as hurricanes. The analysis used data on bottled water purchase behaviors observed in 16 retail chains in affected areas from 4 hurricanes. This joint project with Dr. Benny Martin, Dr. Jun Zhang, and Dr. Martin Dresner, revealed two main insights: one is that supply side, demand side, and disaster characteristics significantly affect stockpiling propensity before hurricanes, which then have lasting impacts on retail performance. The other insight is that drugstores are associated with the highest stockpiling propensity before hurricanes.
The third talk was given by Kyle Hunt, a student and research assistant at University at Buffalo, on “Tracking Misinformation on Twitter During Crisis Events: A Machine Learning Approach.” The presented talk was about a recently concluded work with Puneet Agarwal and Dr. Jun Zhuang. Hunt talked about the tool they developed to help FEMA in tracking the spread of incorrect information on social media regarding natural disasters, so that FEMA or other authorities can respond in time and clarify rumors. When a crisis event like a hurricane happens, the first step is to label social media posts as true or false. Then the tracking component relies on machine learning algorithms, which is trained using the pre-labeled data. Through an extensive numerical study, they showed that the tool can be useful in tracking false news and preventing it from spreading further.
Dr. Elisa Long then concluded the session with her talk “Political Storms: Emergent Partisan Skepticism about Hurricane Risks.” This was a joint work with Dr. Ryne Rohla and Dr. Keith Chen. This research was motivated by the conjectured cultural shift that people with different political views seem to show greater disparity in perceiving hurricane risks, which then affect their evacuation behaviors. Using GPS-based smartphone data collected during Hurricane Matthew, Harvey, and Irma, their work attempts to answer two questions: Did partisan skepticism of hurricane risks affect actual evacuation behavior? What is the causal effect of an official hurricane watch on evacuation behavior? By applying novel analytic techniques on phone data, they concluded that partisan skepticism of hurricane risks does affect actual evacuation behavior, and an official alert can trigger rapid evacuation.